摘要
目的:为了提高阿尔茨海默病(Alzheimer’s disease,AD)分类的准确率,构建一种基于海马MRI数据的DenseNet和通道注意力模块(channel attention module,CAM)相结合的AD分类模型。方法:首先,通过结构磁共振图像提取海马感兴趣区。其次,通过优化网络结构将三维DenseNet与CAM相结合构建基于海马感兴趣区的AD分类模型(三维CAM-DenseNet模型)。最后,为验证该模型的分类性能,将该模型与多个三维DenseNet模型进行比较,并验证加入纵向数据后对模型分类性能的影响;为评估模型的泛化性,将该模型在3个独立测试集上进行检验。结果:三维CAM-DenseNet模型在区分AD患者与认知正常受试者的分类任务中平均准确率为95.2%、敏感度为91.9%、特异度为97.8%、AUC值为94.9%,优于其他三维DenseNet模型;在轻度认知障碍相关分类任务中,加入纵向数据可以提升模型的分类性能;训练好的模型在3个独立测试集中均表现出良好的泛化性能。结论:构建的三维CAM-DenseNet模型分类准确率高、泛化性好,适用于AD分类研究。
Objective To develop a classification model based on hippocampal MRI data using DenseNet and channel attention module(CAM)to enhance the classification accuracy of Alzheimer's disease(AD).Methods Firstly the hippocampal region of interest(ROI)was extracted from structural magnetic resonance images,and then the network structure was optimized to combine 3D DenseNet with CAM so as to establish an Alzheimer's disease classification model based on hippocampal ROI(3D CAM-DenseNet model).The classification model developed was compared with several 3D DenseNet models to verify its performances,and longitudinal data of the subjects involved were added to evaluate their effect on the model;the generaliza-bility of the model was assessed with three independent test sets.Results The 3D CAM-DenseNet model outperformed other 3D DenseNet models with an average accuracy of 95.2%,sensitivity of 91.9%,specificity of 97.8%,and AUC value of 94.9%in distinguishing AD patients from cognitively normal subjects;the involvement of the longitudinal data in tasks retated to mild cognitive impairment classification enhanced the classi-fication performance of the 3D CAM-DenseNet model;the trained 3D CAM-DenseNet model showed high generalizability when used for the three independent test sets.Conclusions The 3D CAM-DenseNet model behaves well in classification accuracy and generalizability,and thus is worthy promoting for AD classification.
作者
金悦
沈小琪
林岚
JIN Yue;SHEN Xiao-qi;LIN Lan(Intelligent Physiological Measurement and Clinical Translation,Beijing International Base for Scientific and Technological Cooperation,Department of Biomedical Engineering,Faculty of Environment and Life,Beijing University of Technology,Beijing 100124,China)
出处
《医疗卫生装备》
CAS
2023年第4期9-14,共6页
Chinese Medical Equipment Journal
基金
国家自然科学基金项目(81971683)
北京市自然科学基金-海淀原始创新联合基金项目(L182010)。